Feature extracting device for pattern recognition

Claim 2

A feature extracting device comprising: a feature vector calculating module for projecting a learning pattern to be recognized on a subspace group, so to calculate squares of projection length on each subspace as feature vectors; and a subspace basis vector learning module including at least a parameter updating module for updating basis vectors of each subspace forming the subspace group, so as to increase the ratio of variation between classes to variation within a class, as for each component of the feature vectors.

Claim 3

A feature extracting device as set forth in claim 1 , in which said feature vector calculating module normalizes the learning pattern, to project the same on the subspace group, and calculates squares of projection length on each subspace, or quantity derived from said squares of projection length, as feature vectors.

Claim 4

A feature extracting device as set forth in claim 1 , in which said subspace basis vector learning module includes a calibrating module for calibrating the feature vectors by performing restraint processing among features based on a restraint parameter predetermined as for the calculated feature vectors.

Claim 5

A feature extracting device as set forth in claim 1 , in which said feature vector calculating normalizes the learning pattern, to project the same on the subspace group, and calculates squares of projection length on each subspace, or quantity derived from there, as feature vectors, and said subspace vector learning module includes a calibrating module for calibrating the feature vectors by performing restraint processing among features based on a restraint parameter predetermined as for the calculated feature vectors.

Claim 6

A feature extracting device as set forth in claim 1 , in which said parameter updating module performs normalized orthogonalization on the basis vectors obtained by update processing according to a Gram-Schmid orthogonalization.

Claim 7

A feature extracting device as set forth in claim 1 , in which said feature vector calculating module normalizes the learning pattern, to project the same on the subspace group, and calculates squares of projection length on each subspace, or quantity derived from said squares of projection length, as feature vectors, and said parameter updating module performs normalized orthogonalization on the basis vectors obtained by update processing, according to a Gram-Schmid orthogonalization.

Claim 8

A feature extracting device as set forth in claim 1 , in which said feature vector calculating module normalizes the learning pattern, to project the same on the subspace group, and calculates squares of projection length on each subspace, or quantity derived from said squares of projection length, as feature vectors, said subspace basis vector learning module includes calibrating module for calibrating the feature vectors by performing restraint processing among features based on a restraint parameter predetermined as for the calculated feature vectors, and said parameter updating module performs normalized orthogonalization on the basis vectors obtained by update processing, according to a Gram-Schmid orthogonalization.

Claim 9

A feature extracting device as set forth in claim 1 , in which said feature vector calculating module normalizes the learning pattern, to project the same on the subspace group, and calculates squares of generalized projection length on each subspace as feature vectors.

Claim 10

A feature extracting device as set forth in claim 1 , in which said subspace basis vector learning unit performs update processing of the basis vectors for increasing the ratio of the variation between classes to the variation within a class as for the feature vectors, by updating the basis vectors so as to make the respective components of the feature vectors have no correlation to each other or make said respective components independent, and simultaneously so as to increase the ratio of the variation between classes to the variation within a class as for the respective components of the feature vectors.

Claim 11

A feature extracting device for deciding features, using a data set, as learning data, comprising input patterns, class names C the patterns belong to, and a series of subclass names C m (m=1 to n, where n is an integer 1 and more, assuming that the subclasses are hierarchically classified finer as m becomes larger), the device by comprising: (n+1) stages of feature extracting layers, a first stage of feature extracting layer comprising a first feature vector calculating module for projecting an input learning pattern, after having been normalized, on a first subspace group, and calculating squares of generalized projection length on each subspace, or quantity derived from said squares of projection length, as first feature vectors; and a first subspace basis vector learning module including at least a first parameter updating module for updating basis vectors of each subspace forming the first subspace group, so as to increase the ratio of variation between subclasses to variation within a subclass as for the n-th subclass, including the most segmented subclass, with respect to the first feature vectors, a k-th (k=2 to n) stage of feature extraction layer comprising: a k-th feature vector calculating module for projecting the (k−1)-th feature vectors calculated in the (k−1)-th stage of feature extraction layer, after having been normalized, on the k-th subspace group, and calculating squares of generalized projection length on each subspace, or quantity derived from said squares of projection length as the k-th feature vectors; and a k-th subspace basis vector learning module including at least a k-th parameter updating module for updating basis vectors of each subspace forming the k-th subspace group, so as to increase the ratio (variation between subclasses/variation within a subclass) as for the (n+1-k)-th subclass, with respect to the k-th feature vectors.

Claim 12

A feature extracting device for deciding features, using a data set, as learning data, comprising input patterns, class names C the patterns belong to, and a series of subclass names C m (m=1 to n, where n is an integer 1 and more, assuming that the subclasses are hierarchically classified finer as m becomes larger), the device by comprising: (n+1) stages of feature extracting layers, a first stage of feature extracting layer comprising a first feature vector calculating module for projecting an input learning pattern, after having been normalized, on a first subspace group, and calculating squares of generalized projection length on each subspace, or quantity derived from said squares of projection length, as first feature vectors; and a first subspace basis vector learning module including at least a first parameter updating module for updating basis vectors of each subspace forming the first subspace group, so as to increase the ratio of variation between subclasses to variation within a subclass as for the n-th subclass, including the most segmented subclass, with respect to the first feature vectors, a k-th (k=2 to n) stage of feature extraction layer comprising: a k-th feature vector calculating module for projecting the (k−1)-th feature vectors calculated in the (k−1)-th stage of feature extraction layer, after having been normalized, on the k-th subspace group, and calculating squares of generalized projection length on each subspace, or quantity derived from said squares of projection length as the k-th feature vectors; and a k-th subspace basis vector learning module including at least a k-th parameter updating module for updating basis vectors of each subspace forming the k-th subspace group, so as to increase the ratio (variation between subclasses/variation within a subclass) as for the (n+1-k)-th subclass, with respect to the k-th feature vectors.

Claim 13

A feature extracting device as set forth in claim 10 , in which an (n+1)-th stage of feature extraction layer comprises: an (n+1)-th feature vector calculating module for projecting the n-th feature vectors calculated in the n-th stage of feature extraction layer, after having been normalized, on the (n+1)-th subspace group, and calculating squares of generalized projection length on each subspace, or quantity derived from said squares of projection length as the k-th feature vectors; and an (n+1)-th subspace basis vector learning module including at least an (n+1)-th parameter updating module for updating basis vectors of each subspace forming the (n+1)-th subspace group, so as to increase the ratio of variation between classes to variation within a class as for the final feature vectors.

Claim 14

A feature extracting device as set forth in claim 10 , in which said subspace basis vector learning module of the respective feature extraction layers includes a calibrating module for calibrating the feature vectors by performing restraint processing among features, based on a restraint parameter predetermined as for the calculated feature vectors.

Claim 15

A feature extracting device as set forth in claim 10 , in which the (n+1)-th stage of feature extraction layer comprises: an (n+1)-th feature vector calculating module for projecting the n-th feature vectors calculated in the n-th stage of feature extraction layer, after having been normalized, on the (n+1)-th subspace group, and calculating squares of generalized projection length on each subspace, or quantity derived from there as the k-th feature vectors; and an (n+1)-th subspace basis vector learning module including at least an (n+1)-th parameter updating module for updating basis vectors of each subspace forming the (n+1)-th subspace group, so as to increase the ratio of variation between classes to variation within a class as for the final feature vectors, and said subspace basis vector learning module of the respective feature extraction layers include calibrating module for calibrating the feature vectors by performing restraint processing among features based on a restraint parameter predetermined as for the calculated feature vectors.

Claim 16

A feature extracting device as set forth in claim 10 , in which said parameter updating module performs normalized orthogonalization on the basis vectors obtained by update processing, according to a Gram-Schmid orthogonalization.

Claim 17

A feature extracting device as set forth in claim 10 , in which said subspace basis vector learning module of the respective feature extraction layers includes a calibrating module for calibrating the feature vectors by performing restraint processing among features based on a restraint parameter predetermined as for the calculated feature vectors, and said parameter updating module performs normalized orthogonalization on the basis vectors obtained by update processing, according to a Gram-Schmid orthogonalization.

Claim 18

A feature extracting device as set forth in claim 10 , in which the (n+1)-th stage of feature extraction layer comprises: an (n+1)-th feature vector calculating module for projecting the n-th feature vectors calculated in the n-th stage of feature extraction layer, after having been normalized, on the (n+1)-th subspace group, and calculating squares of generalized projection length on each subspace, or quantity derived from said squares of projection length as the k-th feature vectors; and an (n+1)-th subspace basis vector learning module including at least an (n+1)-th parameter updating module for updating basis vectors of each subspace forming the (n+1)-th subspace group, so as to increase the ratio of variation between classes to variation within a class as for the final feature vectors, said subspace basis vector learning module of the respective feature extraction layers includes a calibrating module for calibrating the feature vectors by performing restraint processing among features based on a restraint parameter predetermined as for the calculated feature vectors, and said parameter updating module performs normalized orthogonalization on the basis vectors obtained by update processing, according to a Gram-Schmid orthogonalization.

Claim 19

A feature extracting device as set forth in claim 10 , in which said feature vector calculating module of each feature extraction layer normalizes an input to the corresponding layer, projects the same on a subspace group, and calculates squares of projection length on each subspace, or quantity derived from said squares of projection length as feature vectors, and said parameter updating module of each feature extraction layer updates normalized orthogonal basis vectors of each subspace forming the subspace group, so as to increase the ratio of variation between subclasses to variation within a subclass, or the ratio of variation between classes to variation within a class, as for the calculated feature vectors.

Claim 20

A feature extracting device as set forth in claim 10 , in which said hierarchical subspace basis vector learning unit performs update processing of the basis vectors, or the normalized orthogonal basis vectors increasing the ratio of the variation between classes to the variation within a class, or the ratio of the variation between subclasses to the variation within a subclass, as for the feature vectors, by updating the basis vectors, or the normalized orthogonal basis vectors so as to make the respective components of the feature vectors have no correlation to each other or make the same independent, and simultaneously so as to increase the ratio of the variation between classes to the variation within a class, or the ratio of the variation between subclasses to the variation within a subclass, as for the respective components of the feature vectors.

Claim 21

A feature extracting device as set forth in claim 10 , in which the (n+1)-th stage of feature extraction layer comprises: an (n+1)-th feature vector calculating module for projecting the n-th feature vectors calculated in the n-th stage of feature extraction layer, after having been normalized, on the (n+1)-th subspace group, and calculating squares of generalized projection length on each subspace, or quantity derived from there as the k-th feature vectors; an (n+1)-th subspace vector learning module including at least an (n+1)-th parameter updating module for updating basis vectors of each subspace forming the (n+1)-th subspace group, so as to increase the ratio of variation between classes to variation within a class as for the final feature vectors; and a hierarchical subspace basis vector learning unit that performs update processing of the basis vectors or the normalized orthogonal basis vectors, increasing the ratio of the variation between classes to the variation within a class, or the ratio of the variation between subclasses to the variation within a subclass, as for the feature vectors, by updating the basis vectors or the normalized orthogonal basis vectors, so as to make the respective components of the feature vectors have no correlation to each other or make said components independent, and simultaneously so as to increase the ratio of the variation between classes to the variation within a class, or the ratio of the variation between subclasses to the variation within a subclass, as for the respective components of the feature vectors.

Claim 22

A feature extracting device having n (n is an integer more than 1) stages of feature extraction layers and hierarchical subspace basis vector learning modules for updating each parameter for describing operations of the respective feature extraction layers, in which: a first stage of feature extracting layer comprises: a first feature vector calculating module for projecting an input pattern, after having been normalized, on a first subspace group, and calculating squares of generalized projection length on each subspace, or quantity derived from there as first feature vectors; a k-th (k=2 to n) stage of feature extraction layer comprises: a k-th feature vector calculating module for projecting the (k−1)-th feature vectors calculated in the (k-1)-th stage of feature extraction layer, after having been normalized, on the k-th subspace group, and calculating squares of generalized projection length on each subspace, or quantity derived from said squares of projection length as the k-th feature vectors; and said hierarchical subspace basis vector learning module includes a module for updating the basis vectors of each subspace forming the subspace group of the respective feature extraction layers, so as to increase the ratio of variation between classes to the variation within a class as for the n-th feature vectors that are the final feature vectors calculated in the n-th stage of feature extraction layer.

Claim 23

A feature extracting device having n (n is an integer more than 1) stages of feature extraction layers and hierarchical subspace basis vector learning modules for updating each parameter for describing operations of the respective feature extraction layers, in which: a first stage of feature extracting layer comprises: a first feature vector calculating module for projecting an input pattern, after having been normalized, on a first subspace group, and calculating squares of generalized projection length on each subspace, or quantity derived from there as first feature vectors; a k-th (k=2 to n) stage of feature extraction layer comprises: a k-th feature vector calculating module for projecting the (k−1)-th feature vectors calculated in the (k-1)-th stage of feature extraction layer, after having been normalized, on the k-th subspace group, and calculating squares of generalized projection length on each subspace, or quantity derived from said squares of projection length as the k-th feature vectors; and said hierarchical subspace basis vector learning module includes a module for updating the basis vectors of each subspace forming the subspace group of the respective feature extraction layers, so as to increase the ratio of variation between classes to the variation within a class as for the n-th feature vectors that are the final feature vectors calculated in the n-th stage of feature extraction layer.

Claim 24

A feature extracting device as set forth in claim 20 , in which said hierarchical subspace basis vector learning module updates the normalized orthogonal basis vectors of each subspace forming the subspace group of the respective feature extraction layers, so as to increase the ratio of the variation between classes to the variation within a class as for the final feature vectors.

Claim 25

A feature extracting device as set forth in claim 20 , in which said subspace basis vector learning module of the respective features extraction layers include a calibrating module for calibrating the feature vectors by performing restraint processing among features based on a restraint parameter predetermined as for the calculated feature vectors.

Claim 26

A feature extracting device as set forth in claim 20 , in which: said hierarchical subspace basis vector learning module updates the normalized orthogonal basis vectors of each subspace forming the subspace group of the respective feature extraction layers, so as to increase the ratio of the variation between classes to the variation within a class as for the final feature vectors, and said subspace basis vector learning module of the respective feature extraction layers include calibrating module for calibrating the feature vectors by performing restraint processing among features based on a restraint parameter predetermined as for the calculated feature vectors.

Claim 27

A feature extracting device as set forth in claim 20 , in which said parameter updating module performs normalized orthogonalization on the basis vectors obtained by update processing, according to a Gram-Schmid orthogonalization.

Claim 28

A feature extracting device as set forth in claim 20 , in which: said hierarchical subspace basis vector learning module updates the normalized orthogonal basis vectors of each subspace forming the subspace group of the respective feature extraction layers, so as to increase the ratio of the variation between classes to the variation within a class as for the final feature vectors, and said parameter updating module performs normalized orthogonalization on the basis vectors obtained by update processing, according to a Gram-Schmid orthogonalization.

Claim 29

A feature extracting device as set forth in claim 20 , in which said hierarchical subspace basis vector learning module updates the normalized orthogonal basis vectors of each subspace forming the subspace group of the respective feature extraction layers, so as to increase the ratio of the variation between classes to the variation within a class as for the final feature vectors, said subspace basis vector learning module of the respective feature extraction layers includes a calibrating module for calibrating the feature vectors by performing restraint processing among features based on a restraint parameter predetermined as for the calculated feature vectors, and said parameter updating module performs normalized orthogonalization on the basis vectors obtained by update processing, according to a Gram-Schmid orthogonalization.

Claim 30

A feature extracting device as set forth in claim 20 , in which said hierarchical subspace basis vector learning unit performs update processing of the basis vectors, or the normalized orthogonal basis vectors increasing the ratio of the variation between classes to the variation within a class, or the ratio of the variation within a class, or the ratio of the variation between subclasses to the variation within a subclass, as for the feature vectors, by updating the basis vectors, or the normalized orthogonal basis vectors so as to make the respective components of the feature vectors have no correlation to each other or make the same independent, and simultaneously so as to increase the ratio of the variation between classes to the variation within a class, or the ratio of the variation between subclasses to the variation within a subclass, as for the respective components of the feature vectors.

Claim 31

A feature extracting device as set forth in claim 20 , in which said hierarchical subspace basis vector learning module updates the normalized orthogonal basis vectors of each subspace forming the subspace group of the respective feature extraction layers so as to increase the ratio of the variation between classes to the variation within a class as for the final feature vectors, and said hierarchical subspace basis vector learning unit performs update processing of the basis vectors, or the normalized orthogonal basis vectors increasing the ratio of the variation between classes to the variation within a class, or the ratio of the variation between subclasses to the variation within a subclass, as for the feature vectors, by updating the basis vectors, or the normalized orthogonal basis vectors so as to make the respective components of the feature vectors have no correlation to each other or make the same independent, and simultaneously so as to increase the ratio of the variation between classes to the variation within a class, or the ratio of the variation between subclasses to the variation within a subclass, as for the respective components of the feature vectors.

Claim 32

A feature extracting device as set forth in claim 20 , in which: said subspace basis vector learning module of the respective feature extraction layers include calibrating module for calibrating the feature vectors by performing restraint processing among features based on a restraint parameter predetermined as for the calculated feature vectors, and said hierarchical subspace basis vector learning unit performs update processing of the basis vectors, or the normalized orthogonal basis vectors increasing the ratio of the variation between classes to the variation within a class, or the ratio of the variation between subclasses to the variation within a subclass, as for the feature vectors, by updating the basis vectors, or the normalized orthogonal basis vectors so as to make the respective components of the feature vectors have no correlation to each other or make the same independent, and simultaneously so as to increase the ratio of the variation between classes to the variation within a class, or the ratio of the variation between subclasses to the variation within a subclass, as for the respective components of the feature vectors.

Claim 33

A feature extracting device as set forth in claim 20 , in which said hierarchical subspace basis vector learning module updates the normalized orthogonal basis vectors of each subspace forming the subspace group of the respective feature extraction layers, so as to increase the ratio of the variation between classes to the variation within a class as for the final feature vectors, said subspace basis vector learning module of the respective feature extraction layers includes a calibrating module for calibrating the feature vectors by performing restraint processing among features based on a restraint parameter predetermined as for the calculated feature vectors, and said hierarchical subspace basis vector learning unit performs update processing of the basis vectors, or the normalized orthogonal basis vectors increasing the ratio of the variation between classes to the variation within a class, or the ratio of the variation between subclasses to the variation within a subclass, as for the feature vectors, by updating the basis vectors, or the normalized orthogonal basis vectors so as to make the respective components of the feature vectors have no correlation to each other or make the same independent, and simultaneously so as to increase the ratio of the variation between classes to the variation within a class, or the ratio of the variation between subclasses to the variation within a subclass, as for the respective components of the feature vectors.

Claim 34

A pattern learning device for learning relationship between input/output, using a learning data set comprising pairs of an input vector and a desired output vector corresponding to the input vector, said device comprising: n stages (n is an integer 1 and or more) of processing layers; and a parameter updating module for updating each parameter for describing operations of the respective processing layers, a first stage of processing layer comprising a first output calculating module for projecting an input vector, after having been normalized, on a first subspace group, and calculating squares of generalized projection length on each subspace, or quantity derived from said squares of generalized projection length as first output vectors, the a k-th (k=2 to n) stage of processing layer, when n is 2 and more, comprising a k-th output calculating module for projecting the (k−1)-th output vectors calculated in the (k-1)-th stage of processing layer, after having been normalized, on the k-th subspace group, and calculating squares of generalized projection length on each subspace, or quantity derived from said squares of generalized projection length as the k-th output vectors, and said parameter updating module including a module for updating the basis vectors of each subspace of the respective processing layers, so as to decrease the average square error of the n-th output vectors calculated in the n-th stage of processing layer, that are the final output vectors, and desired output vectors corresponding to the input vector.

Claim 35

A pattern learning device for learning relationship between input/output, using a learning data set comprising pairs of an input vector and a desired output vector corresponding to the input vector, said device comprising: n stages (n is an integer 1 and or more) of processing layers; and a parameter updating module for updating each parameter for describing operations of the respective processing layers, a first stage of processing layer comprising a first output calculating module for projecting an input vector, after having been normalized, on a first subspace group, and calculating squares of generalized projection length on each subspace, or quantity derived from said squares of generalized projection length as first output vectors, the a k-th (k=2 to n) stage of processing layer, when n is 2 and more, comprising a k-th output calculating module for projecting the (k−1)-th output vectors calculated in the (k-1)-th stage of processing layer, after having been normalized, on the k-th subspace group, and calculating squares of generalized projection length on each subspace, or quantity derived from said squares of generalized projection length as the k-th output vectors, and said parameter updating module including a module for updating the basis vectors of each subspace of the respective processing layers, so as to decrease the average square error of the n-th output vectors calculated in the n-th stage of processing layer, that are the final output vectors, and desired output vectors corresponding to the input vector.

Claim 36

A computer readable memory storing a feature extraction program for extracting features for pattern recognition, controlling a computer, said feature extraction program performing a method comprising: projecting a learning pattern to be recognized on a subspace group, so as to calculate squares of projection length on each subspace as feature vectors; and updating basis vectors of each subspace forming the subspace group, so as to increase the ratio of variation between classes to variation within a class, as for each component of the feature vectors.

Claim 37

A computer readable memory storing a feature extraction program for extracting features for pattern recognition, controlling a computer, said feature extraction program performing a method comprising: projecting a learning pattern to be recognized on a subspace group, so as to calculate squares of projection length on each subspace as feature vectors; and updating basis vectors of each subspace forming the subspace group, so as to increase the ratio of variation between classes to variation within a class, as for each component of the feature vectors.

Claim 38

A computer readable memory storing the feature extraction program as set forth in claim 32 , the feature extraction program normalizing the learning pattern, to project the same on the subspace group, and calculating squares of generalized projection length on each subspace, as feature vectors, in the feature vector calculation.

Claim 39

A computer readable memory storing the feature extraction program as set forth in claim 32 , the feature extraction program performing update processing of the basis vectors for increasing the ratio of the variation between classes to the variation within a class as for the feature vectors, by updating the basis vectors so as to make the respective components of the feature vectors have no correlation to each other or make said components independent, and simultaneously so as to increase the ratio of the variation between classes to the variation within a class as for the respective components of the feature vectors.

Claim 40

A computer readable memory storing the feature extraction program as set forth in claim 32 , the feature extraction program normalizing the learning pattern, to project the same on the subspace group, and calculating squares of generalized projection length on each subspace, as feature vectors, in the feature vector calculation, and performing update processing of the basis vectors for increasing the ratio of the variation between classes to the variation within a class as for the feature vectors, by updating the basis vectors so as to make the respective components of the feature vectors have no correlation to each other or make said components independent, and simultaneously so as to increase the ratio of the variation between classes to the variation within a class as for the respective components of the feature vectors.

Claim 41

A computer readable memory storing a feature extraction program for deciding features, using data set, as learning data, consisting of input patterns, class names C the patterns belong to, and a series of subclass names C m (m=1 to n, where n is an integer 1 and more, assuming that the subclasses are hierarchically classified finer according as m becomes larger), the feature extraction program comprising: (n+1) stages of feature extraction layers, a first stage of feature extracting layer comprising: a first feature vector calculating module for projecting an input learning pattern, after having been normalized, on a first subspace group, and calculating squares of generalized projection length on each subspace, or quantity derived from said squares of generalized projection length as first feature vectors; and a first subspace basis vector learning module including at least a first parameter updating module for updating basis vectors of each subspace forming the first subspace group, so as to increase the ratio of variation between subclasses to variation within a subclass as for the n-th subclass, including the most segmented subclass, with respect to the first feature vectors, the k-th (k=2 to n) stage of feature extraction layer comprising: a k-th feature vector calculating module for projecting the (k−1)-th feature vectors calculated in the (k−1)-th stage of feature extraction layer, after having been normalized, on the k-th subspace group, and calculating squares of generalized projection length on each subspace, or quantity derived from said squares of generalized projection length as the k-th feature vectors, and a k-th subspace basis vector learning module including at least a k-th parameter updating module for updating basis vectors of each subspace forming the k-th subspace group, so as to increase the ratio (variation between subclasses/variation within a subclass) as for the (n+1-k)-th subclass, with respect to the k-th feature vectors.

Claim 42

A computer readable memory storing a feature extraction program for deciding features, using data set, as learning data, consisting of input patterns, class names C the patterns belong to, and a series of subclass names C m (m=1 to n, where n is an integer 1 and more, assuming that the subclasses are hierarchically classified finer according as m becomes larger), the feature extraction program comprising: (n+1) stages of feature extraction layers, a first stage of feature extracting layer comprising: a first feature vector calculating module for projecting an input learning pattern, after having been normalized, on a first subspace group, and calculating squares of generalized projection length on each subspace, or quantity derived from said squares of generalized projection length as first feature vectors; and a first subspace basis vector learning module including at least a first parameter updating module for updating basis vectors of each subspace forming the first subspace group, so as to increase the ratio of variation between subclasses to variation within a subclass as for the n-th subclass, including the most segmented subclass, with respect to the first feature vectors, the k-th (k=2 to n) stage of feature extraction layer comprising: a k-th feature vector calculating module for projecting the (k−1)-th feature vectors calculated in the (k−1)-th stage of feature extraction layer, after having been normalized, on the k-th subspace group, and calculating squares of generalized projection length on each subspace, or quantity derived from said squares of generalized projection length as the k-th feature vectors, and a k-th subspace basis vector learning module including at least a k-th parameter updating module for updating basis vectors of each subspace forming the k-th subspace group, so as to increase the ratio (variation between subclasses/variation within a subclass) as for the (n+1-k)-th subclass, with respect to the k-th feature vectors.

Claim 43

A computer readable memory storing the feature extraction program as set forth in claim 36 , in which the (n+1)-th stage of feature extraction layer comprises: an (n+1)-th feature vector calculating module for projecting the n-th feature vectors calculated in the n-th stage of feature extraction layer, after having been normalized, on the (n+1)-th subspace group, and calculating squares of generalized projection length on each subspace, or quantity derived from there as the k-th feature vectors, and an (n+1)-th parameter updating module for updating basis vectors of each subspace forming the (n+1)-th subspace group, so as to increase the ratio of variation between classes to variation within a class as for the final feature vectors.

Claim 44

A computer readable memory storing a feature extraction program for realizing n (n is an integer more than 1) stages of feature extraction layers and a hierarchical subspace basis vector learning function for updating each parameter for describing operations of the respective feature extraction layers, the feature extraction program comprising: in a first stage of feature extracting layer, a first feature vector calculating module for projecting an input pattern, after having been normalized on a first subspace group, and calculating squares of generalized projection length on each subspace, or quantity derived therefrom as first feature vectors; and in the each of a k-th (k=2 to n) stage of feature extraction layer, a k-th feature vector calculating module for projecting the (k−1)-th feature vectors calculated in the (k-1)-th stage of feature extraction layer, after having been normalized, on the k-th subspace group, and calculating squares of generalized projection length on each subspace, or quantity derived from there as the k-th feature vectors, and said hierarchical subspace basis vector learning function updating the basis vectors of each subspace forming the subspace groups of the respective feature extraction layers, so as to increase the ratio of the variation between classes to the variation within a class as for the n-th feature vectors that are the final feature vectors calculated in the n-th stage of feature extraction layer.

Claim 45

A computer readable memory storing a feature extraction program for realizing n (n is an integer more than 1) stages of feature extraction layers and a hierarchical subspace basis vector learning function for updating each parameter for describing operations of the respective feature extraction layers, the feature extraction program comprising: in a first stage of feature extracting layer, a first feature vector calculating module for projecting an input pattern, after having been normalized on a first subspace group, and calculating squares of generalized projection length on each subspace, or quantity derived therefrom as first feature vectors; and in the each of a k-th (k=2 to n) stage of feature extraction layer, a k-th feature vector calculating module for projecting the (k−1)-th feature vectors calculated in the (k-1)-th stage of feature extraction layer, after having been normalized, on the k-th subspace group, and calculating squares of generalized projection length on each subspace, or quantity derived from there as the k-th feature vectors, and said hierarchical subspace basis vector learning function updating the basis vectors of each subspace forming the subspace groups of the respective feature extraction layers, so as to increase the ratio of the variation between classes to the variation within a class as for the n-th feature vectors that are the final feature vectors calculated in the n-th stage of feature extraction layer.